We analysed the simulated data using Cox and logistic regression models with genotype and baseline age included as covariates in the model. The Cox model was fitted using the time-on-study time-scale in both the cohort and case-cohort simulations. In the case-cohort setting, Prentice weights and robust SEs were used for the Cox model to account for the sampling process.11, 12 Logistic regression in case-cohort studies applied directly to the cases and the subcohort non-cases gives asymptotic inference of odds ratios.11 Each simulation scenario was repeated 5000 times. The P-values were calculated using Wald tests.
